Commit | Line | Data |
---|---|---|
b8441ed2 TH |
1 | /* |
2 | * fs/kernfs/dir.c - kernfs directory implementation | |
3 | * | |
4 | * Copyright (c) 2001-3 Patrick Mochel | |
5 | * Copyright (c) 2007 SUSE Linux Products GmbH | |
6 | * Copyright (c) 2007, 2013 Tejun Heo <tj@kernel.org> | |
7 | * | |
8 | * This file is released under the GPLv2. | |
9 | */ | |
fd7b9f7b | 10 | |
abd54f02 | 11 | #include <linux/sched.h> |
fd7b9f7b TH |
12 | #include <linux/fs.h> |
13 | #include <linux/namei.h> | |
14 | #include <linux/idr.h> | |
15 | #include <linux/slab.h> | |
16 | #include <linux/security.h> | |
17 | #include <linux/hash.h> | |
18 | ||
19 | #include "kernfs-internal.h" | |
20 | ||
a797bfc3 | 21 | DEFINE_MUTEX(kernfs_mutex); |
3eef34ad TH |
22 | static DEFINE_SPINLOCK(kernfs_rename_lock); /* kn->parent and ->name */ |
23 | static char kernfs_pr_cont_buf[PATH_MAX]; /* protected by rename_lock */ | |
fd7b9f7b | 24 | |
adc5e8b5 | 25 | #define rb_to_kn(X) rb_entry((X), struct kernfs_node, rb) |
fd7b9f7b | 26 | |
81c173cb TH |
27 | static bool kernfs_active(struct kernfs_node *kn) |
28 | { | |
29 | lockdep_assert_held(&kernfs_mutex); | |
30 | return atomic_read(&kn->active) >= 0; | |
31 | } | |
32 | ||
182fd64b TH |
33 | static bool kernfs_lockdep(struct kernfs_node *kn) |
34 | { | |
35 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
36 | return kn->flags & KERNFS_LOCKDEP; | |
37 | #else | |
38 | return false; | |
39 | #endif | |
40 | } | |
41 | ||
3eef34ad TH |
42 | static int kernfs_name_locked(struct kernfs_node *kn, char *buf, size_t buflen) |
43 | { | |
44 | return strlcpy(buf, kn->parent ? kn->name : "/", buflen); | |
45 | } | |
46 | ||
47 | static char * __must_check kernfs_path_locked(struct kernfs_node *kn, char *buf, | |
48 | size_t buflen) | |
49 | { | |
50 | char *p = buf + buflen; | |
51 | int len; | |
52 | ||
53 | *--p = '\0'; | |
54 | ||
55 | do { | |
56 | len = strlen(kn->name); | |
57 | if (p - buf < len + 1) { | |
58 | buf[0] = '\0'; | |
59 | p = NULL; | |
60 | break; | |
61 | } | |
62 | p -= len; | |
63 | memcpy(p, kn->name, len); | |
64 | *--p = '/'; | |
65 | kn = kn->parent; | |
66 | } while (kn && kn->parent); | |
67 | ||
68 | return p; | |
69 | } | |
70 | ||
71 | /** | |
72 | * kernfs_name - obtain the name of a given node | |
73 | * @kn: kernfs_node of interest | |
74 | * @buf: buffer to copy @kn's name into | |
75 | * @buflen: size of @buf | |
76 | * | |
77 | * Copies the name of @kn into @buf of @buflen bytes. The behavior is | |
78 | * similar to strlcpy(). It returns the length of @kn's name and if @buf | |
79 | * isn't long enough, it's filled upto @buflen-1 and nul terminated. | |
80 | * | |
81 | * This function can be called from any context. | |
82 | */ | |
83 | int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen) | |
84 | { | |
85 | unsigned long flags; | |
86 | int ret; | |
87 | ||
88 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
89 | ret = kernfs_name_locked(kn, buf, buflen); | |
90 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
91 | return ret; | |
92 | } | |
93 | ||
94 | /** | |
95 | * kernfs_path - build full path of a given node | |
96 | * @kn: kernfs_node of interest | |
97 | * @buf: buffer to copy @kn's name into | |
98 | * @buflen: size of @buf | |
99 | * | |
100 | * Builds and returns the full path of @kn in @buf of @buflen bytes. The | |
101 | * path is built from the end of @buf so the returned pointer usually | |
102 | * doesn't match @buf. If @buf isn't long enough, @buf is nul terminated | |
103 | * and %NULL is returned. | |
104 | */ | |
105 | char *kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen) | |
106 | { | |
107 | unsigned long flags; | |
108 | char *p; | |
109 | ||
110 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
111 | p = kernfs_path_locked(kn, buf, buflen); | |
112 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
113 | return p; | |
114 | } | |
e61734c5 | 115 | EXPORT_SYMBOL_GPL(kernfs_path); |
3eef34ad TH |
116 | |
117 | /** | |
118 | * pr_cont_kernfs_name - pr_cont name of a kernfs_node | |
119 | * @kn: kernfs_node of interest | |
120 | * | |
121 | * This function can be called from any context. | |
122 | */ | |
123 | void pr_cont_kernfs_name(struct kernfs_node *kn) | |
124 | { | |
125 | unsigned long flags; | |
126 | ||
127 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
128 | ||
129 | kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf)); | |
130 | pr_cont("%s", kernfs_pr_cont_buf); | |
131 | ||
132 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
133 | } | |
134 | ||
135 | /** | |
136 | * pr_cont_kernfs_path - pr_cont path of a kernfs_node | |
137 | * @kn: kernfs_node of interest | |
138 | * | |
139 | * This function can be called from any context. | |
140 | */ | |
141 | void pr_cont_kernfs_path(struct kernfs_node *kn) | |
142 | { | |
143 | unsigned long flags; | |
144 | char *p; | |
145 | ||
146 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
147 | ||
148 | p = kernfs_path_locked(kn, kernfs_pr_cont_buf, | |
149 | sizeof(kernfs_pr_cont_buf)); | |
150 | if (p) | |
151 | pr_cont("%s", p); | |
152 | else | |
153 | pr_cont("<name too long>"); | |
154 | ||
155 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
156 | } | |
157 | ||
158 | /** | |
159 | * kernfs_get_parent - determine the parent node and pin it | |
160 | * @kn: kernfs_node of interest | |
161 | * | |
162 | * Determines @kn's parent, pins and returns it. This function can be | |
163 | * called from any context. | |
164 | */ | |
165 | struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn) | |
166 | { | |
167 | struct kernfs_node *parent; | |
168 | unsigned long flags; | |
169 | ||
170 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
171 | parent = kn->parent; | |
172 | kernfs_get(parent); | |
173 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
174 | ||
175 | return parent; | |
176 | } | |
177 | ||
fd7b9f7b | 178 | /** |
c637b8ac | 179 | * kernfs_name_hash |
fd7b9f7b TH |
180 | * @name: Null terminated string to hash |
181 | * @ns: Namespace tag to hash | |
182 | * | |
183 | * Returns 31 bit hash of ns + name (so it fits in an off_t ) | |
184 | */ | |
c637b8ac | 185 | static unsigned int kernfs_name_hash(const char *name, const void *ns) |
fd7b9f7b TH |
186 | { |
187 | unsigned long hash = init_name_hash(); | |
188 | unsigned int len = strlen(name); | |
189 | while (len--) | |
190 | hash = partial_name_hash(*name++, hash); | |
191 | hash = (end_name_hash(hash) ^ hash_ptr((void *)ns, 31)); | |
192 | hash &= 0x7fffffffU; | |
193 | /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */ | |
88391d49 | 194 | if (hash < 2) |
fd7b9f7b TH |
195 | hash += 2; |
196 | if (hash >= INT_MAX) | |
197 | hash = INT_MAX - 1; | |
198 | return hash; | |
199 | } | |
200 | ||
c637b8ac TH |
201 | static int kernfs_name_compare(unsigned int hash, const char *name, |
202 | const void *ns, const struct kernfs_node *kn) | |
fd7b9f7b | 203 | { |
adc5e8b5 TH |
204 | if (hash != kn->hash) |
205 | return hash - kn->hash; | |
206 | if (ns != kn->ns) | |
207 | return ns - kn->ns; | |
208 | return strcmp(name, kn->name); | |
fd7b9f7b TH |
209 | } |
210 | ||
c637b8ac TH |
211 | static int kernfs_sd_compare(const struct kernfs_node *left, |
212 | const struct kernfs_node *right) | |
fd7b9f7b | 213 | { |
c637b8ac | 214 | return kernfs_name_compare(left->hash, left->name, left->ns, right); |
fd7b9f7b TH |
215 | } |
216 | ||
217 | /** | |
c637b8ac | 218 | * kernfs_link_sibling - link kernfs_node into sibling rbtree |
324a56e1 | 219 | * @kn: kernfs_node of interest |
fd7b9f7b | 220 | * |
324a56e1 | 221 | * Link @kn into its sibling rbtree which starts from |
adc5e8b5 | 222 | * @kn->parent->dir.children. |
fd7b9f7b TH |
223 | * |
224 | * Locking: | |
a797bfc3 | 225 | * mutex_lock(kernfs_mutex) |
fd7b9f7b TH |
226 | * |
227 | * RETURNS: | |
228 | * 0 on susccess -EEXIST on failure. | |
229 | */ | |
c637b8ac | 230 | static int kernfs_link_sibling(struct kernfs_node *kn) |
fd7b9f7b | 231 | { |
adc5e8b5 | 232 | struct rb_node **node = &kn->parent->dir.children.rb_node; |
fd7b9f7b TH |
233 | struct rb_node *parent = NULL; |
234 | ||
df23fc39 | 235 | if (kernfs_type(kn) == KERNFS_DIR) |
adc5e8b5 | 236 | kn->parent->dir.subdirs++; |
fd7b9f7b TH |
237 | |
238 | while (*node) { | |
324a56e1 | 239 | struct kernfs_node *pos; |
fd7b9f7b TH |
240 | int result; |
241 | ||
324a56e1 | 242 | pos = rb_to_kn(*node); |
fd7b9f7b | 243 | parent = *node; |
c637b8ac | 244 | result = kernfs_sd_compare(kn, pos); |
fd7b9f7b | 245 | if (result < 0) |
adc5e8b5 | 246 | node = &pos->rb.rb_left; |
fd7b9f7b | 247 | else if (result > 0) |
adc5e8b5 | 248 | node = &pos->rb.rb_right; |
fd7b9f7b TH |
249 | else |
250 | return -EEXIST; | |
251 | } | |
252 | /* add new node and rebalance the tree */ | |
adc5e8b5 TH |
253 | rb_link_node(&kn->rb, parent, node); |
254 | rb_insert_color(&kn->rb, &kn->parent->dir.children); | |
fd7b9f7b TH |
255 | return 0; |
256 | } | |
257 | ||
258 | /** | |
c637b8ac | 259 | * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree |
324a56e1 | 260 | * @kn: kernfs_node of interest |
fd7b9f7b | 261 | * |
35beab06 TH |
262 | * Try to unlink @kn from its sibling rbtree which starts from |
263 | * kn->parent->dir.children. Returns %true if @kn was actually | |
264 | * removed, %false if @kn wasn't on the rbtree. | |
fd7b9f7b TH |
265 | * |
266 | * Locking: | |
a797bfc3 | 267 | * mutex_lock(kernfs_mutex) |
fd7b9f7b | 268 | */ |
35beab06 | 269 | static bool kernfs_unlink_sibling(struct kernfs_node *kn) |
fd7b9f7b | 270 | { |
35beab06 TH |
271 | if (RB_EMPTY_NODE(&kn->rb)) |
272 | return false; | |
273 | ||
df23fc39 | 274 | if (kernfs_type(kn) == KERNFS_DIR) |
adc5e8b5 | 275 | kn->parent->dir.subdirs--; |
fd7b9f7b | 276 | |
adc5e8b5 | 277 | rb_erase(&kn->rb, &kn->parent->dir.children); |
35beab06 TH |
278 | RB_CLEAR_NODE(&kn->rb); |
279 | return true; | |
fd7b9f7b TH |
280 | } |
281 | ||
282 | /** | |
c637b8ac | 283 | * kernfs_get_active - get an active reference to kernfs_node |
324a56e1 | 284 | * @kn: kernfs_node to get an active reference to |
fd7b9f7b | 285 | * |
324a56e1 | 286 | * Get an active reference of @kn. This function is noop if @kn |
fd7b9f7b TH |
287 | * is NULL. |
288 | * | |
289 | * RETURNS: | |
324a56e1 | 290 | * Pointer to @kn on success, NULL on failure. |
fd7b9f7b | 291 | */ |
c637b8ac | 292 | struct kernfs_node *kernfs_get_active(struct kernfs_node *kn) |
fd7b9f7b | 293 | { |
324a56e1 | 294 | if (unlikely(!kn)) |
fd7b9f7b TH |
295 | return NULL; |
296 | ||
f4b3e631 GKH |
297 | if (!atomic_inc_unless_negative(&kn->active)) |
298 | return NULL; | |
895a068a | 299 | |
182fd64b | 300 | if (kernfs_lockdep(kn)) |
f4b3e631 GKH |
301 | rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_); |
302 | return kn; | |
fd7b9f7b TH |
303 | } |
304 | ||
305 | /** | |
c637b8ac | 306 | * kernfs_put_active - put an active reference to kernfs_node |
324a56e1 | 307 | * @kn: kernfs_node to put an active reference to |
fd7b9f7b | 308 | * |
324a56e1 | 309 | * Put an active reference to @kn. This function is noop if @kn |
fd7b9f7b TH |
310 | * is NULL. |
311 | */ | |
c637b8ac | 312 | void kernfs_put_active(struct kernfs_node *kn) |
fd7b9f7b | 313 | { |
abd54f02 | 314 | struct kernfs_root *root = kernfs_root(kn); |
fd7b9f7b TH |
315 | int v; |
316 | ||
324a56e1 | 317 | if (unlikely(!kn)) |
fd7b9f7b TH |
318 | return; |
319 | ||
182fd64b | 320 | if (kernfs_lockdep(kn)) |
324a56e1 | 321 | rwsem_release(&kn->dep_map, 1, _RET_IP_); |
adc5e8b5 | 322 | v = atomic_dec_return(&kn->active); |
df23fc39 | 323 | if (likely(v != KN_DEACTIVATED_BIAS)) |
fd7b9f7b TH |
324 | return; |
325 | ||
abd54f02 | 326 | wake_up_all(&root->deactivate_waitq); |
fd7b9f7b TH |
327 | } |
328 | ||
329 | /** | |
81c173cb TH |
330 | * kernfs_drain - drain kernfs_node |
331 | * @kn: kernfs_node to drain | |
fd7b9f7b | 332 | * |
81c173cb TH |
333 | * Drain existing usages and nuke all existing mmaps of @kn. Mutiple |
334 | * removers may invoke this function concurrently on @kn and all will | |
335 | * return after draining is complete. | |
fd7b9f7b | 336 | */ |
81c173cb | 337 | static void kernfs_drain(struct kernfs_node *kn) |
35beab06 | 338 | __releases(&kernfs_mutex) __acquires(&kernfs_mutex) |
fd7b9f7b | 339 | { |
abd54f02 | 340 | struct kernfs_root *root = kernfs_root(kn); |
fd7b9f7b | 341 | |
35beab06 | 342 | lockdep_assert_held(&kernfs_mutex); |
81c173cb | 343 | WARN_ON_ONCE(kernfs_active(kn)); |
ea1c472d | 344 | |
35beab06 | 345 | mutex_unlock(&kernfs_mutex); |
abd54f02 | 346 | |
182fd64b | 347 | if (kernfs_lockdep(kn)) { |
35beab06 TH |
348 | rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_); |
349 | if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS) | |
350 | lock_contended(&kn->dep_map, _RET_IP_); | |
351 | } | |
abd54f02 | 352 | |
35beab06 | 353 | /* but everyone should wait for draining */ |
abd54f02 TH |
354 | wait_event(root->deactivate_waitq, |
355 | atomic_read(&kn->active) == KN_DEACTIVATED_BIAS); | |
fd7b9f7b | 356 | |
182fd64b | 357 | if (kernfs_lockdep(kn)) { |
a6607930 TH |
358 | lock_acquired(&kn->dep_map, _RET_IP_); |
359 | rwsem_release(&kn->dep_map, 1, _RET_IP_); | |
360 | } | |
35beab06 | 361 | |
ccf02aaf TH |
362 | kernfs_unmap_bin_file(kn); |
363 | ||
35beab06 | 364 | mutex_lock(&kernfs_mutex); |
fd7b9f7b TH |
365 | } |
366 | ||
fd7b9f7b | 367 | /** |
324a56e1 TH |
368 | * kernfs_get - get a reference count on a kernfs_node |
369 | * @kn: the target kernfs_node | |
fd7b9f7b | 370 | */ |
324a56e1 | 371 | void kernfs_get(struct kernfs_node *kn) |
fd7b9f7b | 372 | { |
324a56e1 | 373 | if (kn) { |
adc5e8b5 TH |
374 | WARN_ON(!atomic_read(&kn->count)); |
375 | atomic_inc(&kn->count); | |
fd7b9f7b TH |
376 | } |
377 | } | |
378 | EXPORT_SYMBOL_GPL(kernfs_get); | |
379 | ||
380 | /** | |
324a56e1 TH |
381 | * kernfs_put - put a reference count on a kernfs_node |
382 | * @kn: the target kernfs_node | |
fd7b9f7b | 383 | * |
324a56e1 | 384 | * Put a reference count of @kn and destroy it if it reached zero. |
fd7b9f7b | 385 | */ |
324a56e1 | 386 | void kernfs_put(struct kernfs_node *kn) |
fd7b9f7b | 387 | { |
324a56e1 | 388 | struct kernfs_node *parent; |
ba7443bc | 389 | struct kernfs_root *root; |
fd7b9f7b | 390 | |
adc5e8b5 | 391 | if (!kn || !atomic_dec_and_test(&kn->count)) |
fd7b9f7b | 392 | return; |
324a56e1 | 393 | root = kernfs_root(kn); |
fd7b9f7b | 394 | repeat: |
81c173cb TH |
395 | /* |
396 | * Moving/renaming is always done while holding reference. | |
adc5e8b5 | 397 | * kn->parent won't change beneath us. |
fd7b9f7b | 398 | */ |
adc5e8b5 | 399 | parent = kn->parent; |
fd7b9f7b | 400 | |
81c173cb TH |
401 | WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS, |
402 | "kernfs_put: %s/%s: released with incorrect active_ref %d\n", | |
403 | parent ? parent->name : "", kn->name, atomic_read(&kn->active)); | |
324a56e1 | 404 | |
df23fc39 | 405 | if (kernfs_type(kn) == KERNFS_LINK) |
adc5e8b5 | 406 | kernfs_put(kn->symlink.target_kn); |
2063d608 | 407 | if (!(kn->flags & KERNFS_STATIC_NAME)) |
adc5e8b5 TH |
408 | kfree(kn->name); |
409 | if (kn->iattr) { | |
410 | if (kn->iattr->ia_secdata) | |
411 | security_release_secctx(kn->iattr->ia_secdata, | |
412 | kn->iattr->ia_secdata_len); | |
413 | simple_xattrs_free(&kn->iattr->xattrs); | |
2322392b | 414 | } |
adc5e8b5 TH |
415 | kfree(kn->iattr); |
416 | ida_simple_remove(&root->ino_ida, kn->ino); | |
a797bfc3 | 417 | kmem_cache_free(kernfs_node_cache, kn); |
fd7b9f7b | 418 | |
324a56e1 TH |
419 | kn = parent; |
420 | if (kn) { | |
adc5e8b5 | 421 | if (atomic_dec_and_test(&kn->count)) |
ba7443bc TH |
422 | goto repeat; |
423 | } else { | |
324a56e1 | 424 | /* just released the root kn, free @root too */ |
bc755553 | 425 | ida_destroy(&root->ino_ida); |
ba7443bc TH |
426 | kfree(root); |
427 | } | |
fd7b9f7b TH |
428 | } |
429 | EXPORT_SYMBOL_GPL(kernfs_put); | |
430 | ||
c637b8ac | 431 | static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags) |
fd7b9f7b | 432 | { |
324a56e1 | 433 | struct kernfs_node *kn; |
fd7b9f7b TH |
434 | |
435 | if (flags & LOOKUP_RCU) | |
436 | return -ECHILD; | |
437 | ||
19bbb926 TH |
438 | /* Always perform fresh lookup for negatives */ |
439 | if (!dentry->d_inode) | |
440 | goto out_bad_unlocked; | |
441 | ||
324a56e1 | 442 | kn = dentry->d_fsdata; |
a797bfc3 | 443 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 444 | |
81c173cb TH |
445 | /* The kernfs node has been deactivated */ |
446 | if (!kernfs_active(kn)) | |
fd7b9f7b TH |
447 | goto out_bad; |
448 | ||
c637b8ac | 449 | /* The kernfs node has been moved? */ |
adc5e8b5 | 450 | if (dentry->d_parent->d_fsdata != kn->parent) |
fd7b9f7b TH |
451 | goto out_bad; |
452 | ||
c637b8ac | 453 | /* The kernfs node has been renamed */ |
adc5e8b5 | 454 | if (strcmp(dentry->d_name.name, kn->name) != 0) |
fd7b9f7b TH |
455 | goto out_bad; |
456 | ||
c637b8ac | 457 | /* The kernfs node has been moved to a different namespace */ |
adc5e8b5 | 458 | if (kn->parent && kernfs_ns_enabled(kn->parent) && |
c525aadd | 459 | kernfs_info(dentry->d_sb)->ns != kn->ns) |
fd7b9f7b TH |
460 | goto out_bad; |
461 | ||
a797bfc3 | 462 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
463 | out_valid: |
464 | return 1; | |
465 | out_bad: | |
a797bfc3 | 466 | mutex_unlock(&kernfs_mutex); |
19bbb926 TH |
467 | out_bad_unlocked: |
468 | /* | |
469 | * @dentry doesn't match the underlying kernfs node, drop the | |
470 | * dentry and force lookup. If we have submounts we must allow the | |
471 | * vfs caches to lie about the state of the filesystem to prevent | |
472 | * leaks and other nasty things, so use check_submounts_and_drop() | |
473 | * instead of d_drop(). | |
fd7b9f7b TH |
474 | */ |
475 | if (check_submounts_and_drop(dentry) != 0) | |
476 | goto out_valid; | |
477 | ||
478 | return 0; | |
479 | } | |
480 | ||
c637b8ac | 481 | static void kernfs_dop_release(struct dentry *dentry) |
fd7b9f7b TH |
482 | { |
483 | kernfs_put(dentry->d_fsdata); | |
484 | } | |
485 | ||
a797bfc3 | 486 | const struct dentry_operations kernfs_dops = { |
c637b8ac | 487 | .d_revalidate = kernfs_dop_revalidate, |
c637b8ac | 488 | .d_release = kernfs_dop_release, |
fd7b9f7b TH |
489 | }; |
490 | ||
0c23b225 TH |
491 | /** |
492 | * kernfs_node_from_dentry - determine kernfs_node associated with a dentry | |
493 | * @dentry: the dentry in question | |
494 | * | |
495 | * Return the kernfs_node associated with @dentry. If @dentry is not a | |
496 | * kernfs one, %NULL is returned. | |
497 | * | |
498 | * While the returned kernfs_node will stay accessible as long as @dentry | |
499 | * is accessible, the returned node can be in any state and the caller is | |
500 | * fully responsible for determining what's accessible. | |
501 | */ | |
502 | struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry) | |
503 | { | |
f41c5934 | 504 | if (dentry->d_sb->s_op == &kernfs_sops) |
0c23b225 TH |
505 | return dentry->d_fsdata; |
506 | return NULL; | |
507 | } | |
508 | ||
db4aad20 TH |
509 | static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root, |
510 | const char *name, umode_t mode, | |
511 | unsigned flags) | |
fd7b9f7b TH |
512 | { |
513 | char *dup_name = NULL; | |
324a56e1 | 514 | struct kernfs_node *kn; |
bc755553 | 515 | int ret; |
fd7b9f7b | 516 | |
2063d608 | 517 | if (!(flags & KERNFS_STATIC_NAME)) { |
fd7b9f7b TH |
518 | name = dup_name = kstrdup(name, GFP_KERNEL); |
519 | if (!name) | |
520 | return NULL; | |
521 | } | |
522 | ||
a797bfc3 | 523 | kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL); |
324a56e1 | 524 | if (!kn) |
fd7b9f7b TH |
525 | goto err_out1; |
526 | ||
bc755553 TH |
527 | ret = ida_simple_get(&root->ino_ida, 1, 0, GFP_KERNEL); |
528 | if (ret < 0) | |
fd7b9f7b | 529 | goto err_out2; |
adc5e8b5 | 530 | kn->ino = ret; |
fd7b9f7b | 531 | |
adc5e8b5 | 532 | atomic_set(&kn->count, 1); |
81c173cb | 533 | atomic_set(&kn->active, KN_DEACTIVATED_BIAS); |
35beab06 | 534 | RB_CLEAR_NODE(&kn->rb); |
fd7b9f7b | 535 | |
adc5e8b5 TH |
536 | kn->name = name; |
537 | kn->mode = mode; | |
81c173cb | 538 | kn->flags = flags; |
fd7b9f7b | 539 | |
324a56e1 | 540 | return kn; |
fd7b9f7b TH |
541 | |
542 | err_out2: | |
a797bfc3 | 543 | kmem_cache_free(kernfs_node_cache, kn); |
fd7b9f7b TH |
544 | err_out1: |
545 | kfree(dup_name); | |
546 | return NULL; | |
547 | } | |
548 | ||
db4aad20 TH |
549 | struct kernfs_node *kernfs_new_node(struct kernfs_node *parent, |
550 | const char *name, umode_t mode, | |
551 | unsigned flags) | |
552 | { | |
553 | struct kernfs_node *kn; | |
554 | ||
555 | kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags); | |
556 | if (kn) { | |
557 | kernfs_get(parent); | |
558 | kn->parent = parent; | |
559 | } | |
560 | return kn; | |
561 | } | |
562 | ||
fd7b9f7b | 563 | /** |
c637b8ac | 564 | * kernfs_add_one - add kernfs_node to parent without warning |
324a56e1 | 565 | * @kn: kernfs_node to be added |
fd7b9f7b | 566 | * |
db4aad20 TH |
567 | * The caller must already have initialized @kn->parent. This |
568 | * function increments nlink of the parent's inode if @kn is a | |
569 | * directory and link into the children list of the parent. | |
fd7b9f7b | 570 | * |
fd7b9f7b TH |
571 | * RETURNS: |
572 | * 0 on success, -EEXIST if entry with the given name already | |
573 | * exists. | |
574 | */ | |
988cd7af | 575 | int kernfs_add_one(struct kernfs_node *kn) |
fd7b9f7b | 576 | { |
db4aad20 | 577 | struct kernfs_node *parent = kn->parent; |
c525aadd | 578 | struct kernfs_iattrs *ps_iattr; |
988cd7af | 579 | bool has_ns; |
fd7b9f7b TH |
580 | int ret; |
581 | ||
988cd7af TH |
582 | mutex_lock(&kernfs_mutex); |
583 | ||
584 | ret = -EINVAL; | |
585 | has_ns = kernfs_ns_enabled(parent); | |
586 | if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", | |
587 | has_ns ? "required" : "invalid", parent->name, kn->name)) | |
588 | goto out_unlock; | |
fd7b9f7b | 589 | |
df23fc39 | 590 | if (kernfs_type(parent) != KERNFS_DIR) |
988cd7af | 591 | goto out_unlock; |
fd7b9f7b | 592 | |
988cd7af | 593 | ret = -ENOENT; |
d35258ef | 594 | if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent)) |
988cd7af | 595 | goto out_unlock; |
798c75a0 | 596 | |
c637b8ac | 597 | kn->hash = kernfs_name_hash(kn->name, kn->ns); |
fd7b9f7b | 598 | |
c637b8ac | 599 | ret = kernfs_link_sibling(kn); |
fd7b9f7b | 600 | if (ret) |
988cd7af | 601 | goto out_unlock; |
fd7b9f7b TH |
602 | |
603 | /* Update timestamps on the parent */ | |
adc5e8b5 | 604 | ps_iattr = parent->iattr; |
fd7b9f7b TH |
605 | if (ps_iattr) { |
606 | struct iattr *ps_iattrs = &ps_iattr->ia_iattr; | |
607 | ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME; | |
608 | } | |
609 | ||
d35258ef TH |
610 | mutex_unlock(&kernfs_mutex); |
611 | ||
612 | /* | |
613 | * Activate the new node unless CREATE_DEACTIVATED is requested. | |
614 | * If not activated here, the kernfs user is responsible for | |
615 | * activating the node with kernfs_activate(). A node which hasn't | |
616 | * been activated is not visible to userland and its removal won't | |
617 | * trigger deactivation. | |
618 | */ | |
619 | if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) | |
620 | kernfs_activate(kn); | |
621 | return 0; | |
622 | ||
988cd7af | 623 | out_unlock: |
a797bfc3 | 624 | mutex_unlock(&kernfs_mutex); |
988cd7af | 625 | return ret; |
fd7b9f7b TH |
626 | } |
627 | ||
628 | /** | |
324a56e1 TH |
629 | * kernfs_find_ns - find kernfs_node with the given name |
630 | * @parent: kernfs_node to search under | |
fd7b9f7b TH |
631 | * @name: name to look for |
632 | * @ns: the namespace tag to use | |
633 | * | |
324a56e1 TH |
634 | * Look for kernfs_node with name @name under @parent. Returns pointer to |
635 | * the found kernfs_node on success, %NULL on failure. | |
fd7b9f7b | 636 | */ |
324a56e1 TH |
637 | static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent, |
638 | const unsigned char *name, | |
639 | const void *ns) | |
fd7b9f7b | 640 | { |
adc5e8b5 | 641 | struct rb_node *node = parent->dir.children.rb_node; |
ac9bba03 | 642 | bool has_ns = kernfs_ns_enabled(parent); |
fd7b9f7b TH |
643 | unsigned int hash; |
644 | ||
a797bfc3 | 645 | lockdep_assert_held(&kernfs_mutex); |
fd7b9f7b TH |
646 | |
647 | if (has_ns != (bool)ns) { | |
c637b8ac | 648 | WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", |
adc5e8b5 | 649 | has_ns ? "required" : "invalid", parent->name, name); |
fd7b9f7b TH |
650 | return NULL; |
651 | } | |
652 | ||
c637b8ac | 653 | hash = kernfs_name_hash(name, ns); |
fd7b9f7b | 654 | while (node) { |
324a56e1 | 655 | struct kernfs_node *kn; |
fd7b9f7b TH |
656 | int result; |
657 | ||
324a56e1 | 658 | kn = rb_to_kn(node); |
c637b8ac | 659 | result = kernfs_name_compare(hash, name, ns, kn); |
fd7b9f7b TH |
660 | if (result < 0) |
661 | node = node->rb_left; | |
662 | else if (result > 0) | |
663 | node = node->rb_right; | |
664 | else | |
324a56e1 | 665 | return kn; |
fd7b9f7b TH |
666 | } |
667 | return NULL; | |
668 | } | |
669 | ||
670 | /** | |
324a56e1 TH |
671 | * kernfs_find_and_get_ns - find and get kernfs_node with the given name |
672 | * @parent: kernfs_node to search under | |
fd7b9f7b TH |
673 | * @name: name to look for |
674 | * @ns: the namespace tag to use | |
675 | * | |
324a56e1 | 676 | * Look for kernfs_node with name @name under @parent and get a reference |
fd7b9f7b | 677 | * if found. This function may sleep and returns pointer to the found |
324a56e1 | 678 | * kernfs_node on success, %NULL on failure. |
fd7b9f7b | 679 | */ |
324a56e1 TH |
680 | struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent, |
681 | const char *name, const void *ns) | |
fd7b9f7b | 682 | { |
324a56e1 | 683 | struct kernfs_node *kn; |
fd7b9f7b | 684 | |
a797bfc3 | 685 | mutex_lock(&kernfs_mutex); |
324a56e1 TH |
686 | kn = kernfs_find_ns(parent, name, ns); |
687 | kernfs_get(kn); | |
a797bfc3 | 688 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b | 689 | |
324a56e1 | 690 | return kn; |
fd7b9f7b TH |
691 | } |
692 | EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns); | |
693 | ||
ba7443bc TH |
694 | /** |
695 | * kernfs_create_root - create a new kernfs hierarchy | |
90c07c89 | 696 | * @scops: optional syscall operations for the hierarchy |
d35258ef | 697 | * @flags: KERNFS_ROOT_* flags |
ba7443bc TH |
698 | * @priv: opaque data associated with the new directory |
699 | * | |
700 | * Returns the root of the new hierarchy on success, ERR_PTR() value on | |
701 | * failure. | |
702 | */ | |
90c07c89 | 703 | struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops, |
d35258ef | 704 | unsigned int flags, void *priv) |
ba7443bc TH |
705 | { |
706 | struct kernfs_root *root; | |
324a56e1 | 707 | struct kernfs_node *kn; |
ba7443bc TH |
708 | |
709 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
710 | if (!root) | |
711 | return ERR_PTR(-ENOMEM); | |
712 | ||
bc755553 TH |
713 | ida_init(&root->ino_ida); |
714 | ||
db4aad20 TH |
715 | kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO, |
716 | KERNFS_DIR); | |
324a56e1 | 717 | if (!kn) { |
bc755553 | 718 | ida_destroy(&root->ino_ida); |
ba7443bc TH |
719 | kfree(root); |
720 | return ERR_PTR(-ENOMEM); | |
721 | } | |
722 | ||
324a56e1 | 723 | kn->priv = priv; |
adc5e8b5 | 724 | kn->dir.root = root; |
ba7443bc | 725 | |
90c07c89 | 726 | root->syscall_ops = scops; |
d35258ef | 727 | root->flags = flags; |
324a56e1 | 728 | root->kn = kn; |
abd54f02 | 729 | init_waitqueue_head(&root->deactivate_waitq); |
ba7443bc | 730 | |
d35258ef TH |
731 | if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) |
732 | kernfs_activate(kn); | |
733 | ||
ba7443bc TH |
734 | return root; |
735 | } | |
736 | ||
737 | /** | |
738 | * kernfs_destroy_root - destroy a kernfs hierarchy | |
739 | * @root: root of the hierarchy to destroy | |
740 | * | |
741 | * Destroy the hierarchy anchored at @root by removing all existing | |
742 | * directories and destroying @root. | |
743 | */ | |
744 | void kernfs_destroy_root(struct kernfs_root *root) | |
745 | { | |
324a56e1 | 746 | kernfs_remove(root->kn); /* will also free @root */ |
ba7443bc TH |
747 | } |
748 | ||
fd7b9f7b TH |
749 | /** |
750 | * kernfs_create_dir_ns - create a directory | |
751 | * @parent: parent in which to create a new directory | |
752 | * @name: name of the new directory | |
bb8b9d09 | 753 | * @mode: mode of the new directory |
fd7b9f7b TH |
754 | * @priv: opaque data associated with the new directory |
755 | * @ns: optional namespace tag of the directory | |
756 | * | |
757 | * Returns the created node on success, ERR_PTR() value on failure. | |
758 | */ | |
324a56e1 | 759 | struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent, |
bb8b9d09 TH |
760 | const char *name, umode_t mode, |
761 | void *priv, const void *ns) | |
fd7b9f7b | 762 | { |
324a56e1 | 763 | struct kernfs_node *kn; |
fd7b9f7b TH |
764 | int rc; |
765 | ||
766 | /* allocate */ | |
db4aad20 | 767 | kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR); |
324a56e1 | 768 | if (!kn) |
fd7b9f7b TH |
769 | return ERR_PTR(-ENOMEM); |
770 | ||
adc5e8b5 TH |
771 | kn->dir.root = parent->dir.root; |
772 | kn->ns = ns; | |
324a56e1 | 773 | kn->priv = priv; |
fd7b9f7b TH |
774 | |
775 | /* link in */ | |
988cd7af | 776 | rc = kernfs_add_one(kn); |
fd7b9f7b | 777 | if (!rc) |
324a56e1 | 778 | return kn; |
fd7b9f7b | 779 | |
324a56e1 | 780 | kernfs_put(kn); |
fd7b9f7b TH |
781 | return ERR_PTR(rc); |
782 | } | |
783 | ||
c637b8ac TH |
784 | static struct dentry *kernfs_iop_lookup(struct inode *dir, |
785 | struct dentry *dentry, | |
786 | unsigned int flags) | |
fd7b9f7b | 787 | { |
19bbb926 | 788 | struct dentry *ret; |
324a56e1 TH |
789 | struct kernfs_node *parent = dentry->d_parent->d_fsdata; |
790 | struct kernfs_node *kn; | |
fd7b9f7b TH |
791 | struct inode *inode; |
792 | const void *ns = NULL; | |
793 | ||
a797bfc3 | 794 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 795 | |
324a56e1 | 796 | if (kernfs_ns_enabled(parent)) |
c525aadd | 797 | ns = kernfs_info(dir->i_sb)->ns; |
fd7b9f7b | 798 | |
324a56e1 | 799 | kn = kernfs_find_ns(parent, dentry->d_name.name, ns); |
fd7b9f7b TH |
800 | |
801 | /* no such entry */ | |
b9c9dad0 | 802 | if (!kn || !kernfs_active(kn)) { |
19bbb926 | 803 | ret = NULL; |
fd7b9f7b TH |
804 | goto out_unlock; |
805 | } | |
324a56e1 TH |
806 | kernfs_get(kn); |
807 | dentry->d_fsdata = kn; | |
fd7b9f7b TH |
808 | |
809 | /* attach dentry and inode */ | |
c637b8ac | 810 | inode = kernfs_get_inode(dir->i_sb, kn); |
fd7b9f7b TH |
811 | if (!inode) { |
812 | ret = ERR_PTR(-ENOMEM); | |
813 | goto out_unlock; | |
814 | } | |
815 | ||
816 | /* instantiate and hash dentry */ | |
817 | ret = d_materialise_unique(dentry, inode); | |
818 | out_unlock: | |
a797bfc3 | 819 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
820 | return ret; |
821 | } | |
822 | ||
80b9bbef TH |
823 | static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry, |
824 | umode_t mode) | |
825 | { | |
826 | struct kernfs_node *parent = dir->i_private; | |
90c07c89 | 827 | struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops; |
07c7530d | 828 | int ret; |
80b9bbef | 829 | |
90c07c89 | 830 | if (!scops || !scops->mkdir) |
80b9bbef TH |
831 | return -EPERM; |
832 | ||
07c7530d TH |
833 | if (!kernfs_get_active(parent)) |
834 | return -ENODEV; | |
835 | ||
90c07c89 | 836 | ret = scops->mkdir(parent, dentry->d_name.name, mode); |
07c7530d TH |
837 | |
838 | kernfs_put_active(parent); | |
839 | return ret; | |
80b9bbef TH |
840 | } |
841 | ||
842 | static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry) | |
843 | { | |
844 | struct kernfs_node *kn = dentry->d_fsdata; | |
90c07c89 | 845 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
07c7530d | 846 | int ret; |
80b9bbef | 847 | |
90c07c89 | 848 | if (!scops || !scops->rmdir) |
80b9bbef TH |
849 | return -EPERM; |
850 | ||
07c7530d TH |
851 | if (!kernfs_get_active(kn)) |
852 | return -ENODEV; | |
853 | ||
90c07c89 | 854 | ret = scops->rmdir(kn); |
07c7530d TH |
855 | |
856 | kernfs_put_active(kn); | |
857 | return ret; | |
80b9bbef TH |
858 | } |
859 | ||
860 | static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry, | |
861 | struct inode *new_dir, struct dentry *new_dentry) | |
862 | { | |
863 | struct kernfs_node *kn = old_dentry->d_fsdata; | |
864 | struct kernfs_node *new_parent = new_dir->i_private; | |
90c07c89 | 865 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
07c7530d | 866 | int ret; |
80b9bbef | 867 | |
90c07c89 | 868 | if (!scops || !scops->rename) |
80b9bbef TH |
869 | return -EPERM; |
870 | ||
07c7530d TH |
871 | if (!kernfs_get_active(kn)) |
872 | return -ENODEV; | |
873 | ||
874 | if (!kernfs_get_active(new_parent)) { | |
875 | kernfs_put_active(kn); | |
876 | return -ENODEV; | |
877 | } | |
878 | ||
90c07c89 | 879 | ret = scops->rename(kn, new_parent, new_dentry->d_name.name); |
07c7530d TH |
880 | |
881 | kernfs_put_active(new_parent); | |
882 | kernfs_put_active(kn); | |
883 | return ret; | |
80b9bbef TH |
884 | } |
885 | ||
a797bfc3 | 886 | const struct inode_operations kernfs_dir_iops = { |
c637b8ac TH |
887 | .lookup = kernfs_iop_lookup, |
888 | .permission = kernfs_iop_permission, | |
889 | .setattr = kernfs_iop_setattr, | |
890 | .getattr = kernfs_iop_getattr, | |
891 | .setxattr = kernfs_iop_setxattr, | |
892 | .removexattr = kernfs_iop_removexattr, | |
893 | .getxattr = kernfs_iop_getxattr, | |
894 | .listxattr = kernfs_iop_listxattr, | |
80b9bbef TH |
895 | |
896 | .mkdir = kernfs_iop_mkdir, | |
897 | .rmdir = kernfs_iop_rmdir, | |
898 | .rename = kernfs_iop_rename, | |
fd7b9f7b TH |
899 | }; |
900 | ||
c637b8ac | 901 | static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos) |
fd7b9f7b | 902 | { |
324a56e1 | 903 | struct kernfs_node *last; |
fd7b9f7b TH |
904 | |
905 | while (true) { | |
906 | struct rb_node *rbn; | |
907 | ||
908 | last = pos; | |
909 | ||
df23fc39 | 910 | if (kernfs_type(pos) != KERNFS_DIR) |
fd7b9f7b TH |
911 | break; |
912 | ||
adc5e8b5 | 913 | rbn = rb_first(&pos->dir.children); |
fd7b9f7b TH |
914 | if (!rbn) |
915 | break; | |
916 | ||
324a56e1 | 917 | pos = rb_to_kn(rbn); |
fd7b9f7b TH |
918 | } |
919 | ||
920 | return last; | |
921 | } | |
922 | ||
923 | /** | |
c637b8ac | 924 | * kernfs_next_descendant_post - find the next descendant for post-order walk |
fd7b9f7b | 925 | * @pos: the current position (%NULL to initiate traversal) |
324a56e1 | 926 | * @root: kernfs_node whose descendants to walk |
fd7b9f7b TH |
927 | * |
928 | * Find the next descendant to visit for post-order traversal of @root's | |
929 | * descendants. @root is included in the iteration and the last node to be | |
930 | * visited. | |
931 | */ | |
c637b8ac TH |
932 | static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos, |
933 | struct kernfs_node *root) | |
fd7b9f7b TH |
934 | { |
935 | struct rb_node *rbn; | |
936 | ||
a797bfc3 | 937 | lockdep_assert_held(&kernfs_mutex); |
fd7b9f7b TH |
938 | |
939 | /* if first iteration, visit leftmost descendant which may be root */ | |
940 | if (!pos) | |
c637b8ac | 941 | return kernfs_leftmost_descendant(root); |
fd7b9f7b TH |
942 | |
943 | /* if we visited @root, we're done */ | |
944 | if (pos == root) | |
945 | return NULL; | |
946 | ||
947 | /* if there's an unvisited sibling, visit its leftmost descendant */ | |
adc5e8b5 | 948 | rbn = rb_next(&pos->rb); |
fd7b9f7b | 949 | if (rbn) |
c637b8ac | 950 | return kernfs_leftmost_descendant(rb_to_kn(rbn)); |
fd7b9f7b TH |
951 | |
952 | /* no sibling left, visit parent */ | |
adc5e8b5 | 953 | return pos->parent; |
fd7b9f7b TH |
954 | } |
955 | ||
d35258ef TH |
956 | /** |
957 | * kernfs_activate - activate a node which started deactivated | |
958 | * @kn: kernfs_node whose subtree is to be activated | |
959 | * | |
960 | * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node | |
961 | * needs to be explicitly activated. A node which hasn't been activated | |
962 | * isn't visible to userland and deactivation is skipped during its | |
963 | * removal. This is useful to construct atomic init sequences where | |
964 | * creation of multiple nodes should either succeed or fail atomically. | |
965 | * | |
966 | * The caller is responsible for ensuring that this function is not called | |
967 | * after kernfs_remove*() is invoked on @kn. | |
968 | */ | |
969 | void kernfs_activate(struct kernfs_node *kn) | |
970 | { | |
971 | struct kernfs_node *pos; | |
972 | ||
973 | mutex_lock(&kernfs_mutex); | |
974 | ||
975 | pos = NULL; | |
976 | while ((pos = kernfs_next_descendant_post(pos, kn))) { | |
977 | if (!pos || (pos->flags & KERNFS_ACTIVATED)) | |
978 | continue; | |
979 | ||
980 | WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb)); | |
981 | WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS); | |
982 | ||
983 | atomic_sub(KN_DEACTIVATED_BIAS, &pos->active); | |
984 | pos->flags |= KERNFS_ACTIVATED; | |
985 | } | |
986 | ||
987 | mutex_unlock(&kernfs_mutex); | |
988 | } | |
989 | ||
988cd7af | 990 | static void __kernfs_remove(struct kernfs_node *kn) |
fd7b9f7b | 991 | { |
35beab06 TH |
992 | struct kernfs_node *pos; |
993 | ||
994 | lockdep_assert_held(&kernfs_mutex); | |
fd7b9f7b | 995 | |
6b0afc2a TH |
996 | /* |
997 | * Short-circuit if non-root @kn has already finished removal. | |
998 | * This is for kernfs_remove_self() which plays with active ref | |
999 | * after removal. | |
1000 | */ | |
1001 | if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb))) | |
ce9b499c GKH |
1002 | return; |
1003 | ||
c637b8ac | 1004 | pr_debug("kernfs %s: removing\n", kn->name); |
fd7b9f7b | 1005 | |
81c173cb | 1006 | /* prevent any new usage under @kn by deactivating all nodes */ |
35beab06 TH |
1007 | pos = NULL; |
1008 | while ((pos = kernfs_next_descendant_post(pos, kn))) | |
81c173cb TH |
1009 | if (kernfs_active(pos)) |
1010 | atomic_add(KN_DEACTIVATED_BIAS, &pos->active); | |
35beab06 TH |
1011 | |
1012 | /* deactivate and unlink the subtree node-by-node */ | |
fd7b9f7b | 1013 | do { |
35beab06 TH |
1014 | pos = kernfs_leftmost_descendant(kn); |
1015 | ||
1016 | /* | |
81c173cb TH |
1017 | * kernfs_drain() drops kernfs_mutex temporarily and @pos's |
1018 | * base ref could have been put by someone else by the time | |
1019 | * the function returns. Make sure it doesn't go away | |
1020 | * underneath us. | |
35beab06 TH |
1021 | */ |
1022 | kernfs_get(pos); | |
1023 | ||
d35258ef TH |
1024 | /* |
1025 | * Drain iff @kn was activated. This avoids draining and | |
1026 | * its lockdep annotations for nodes which have never been | |
1027 | * activated and allows embedding kernfs_remove() in create | |
1028 | * error paths without worrying about draining. | |
1029 | */ | |
1030 | if (kn->flags & KERNFS_ACTIVATED) | |
1031 | kernfs_drain(pos); | |
1032 | else | |
1033 | WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS); | |
35beab06 TH |
1034 | |
1035 | /* | |
1036 | * kernfs_unlink_sibling() succeeds once per node. Use it | |
1037 | * to decide who's responsible for cleanups. | |
1038 | */ | |
1039 | if (!pos->parent || kernfs_unlink_sibling(pos)) { | |
1040 | struct kernfs_iattrs *ps_iattr = | |
1041 | pos->parent ? pos->parent->iattr : NULL; | |
1042 | ||
1043 | /* update timestamps on the parent */ | |
1044 | if (ps_iattr) { | |
1045 | ps_iattr->ia_iattr.ia_ctime = CURRENT_TIME; | |
1046 | ps_iattr->ia_iattr.ia_mtime = CURRENT_TIME; | |
1047 | } | |
1048 | ||
988cd7af | 1049 | kernfs_put(pos); |
35beab06 TH |
1050 | } |
1051 | ||
1052 | kernfs_put(pos); | |
1053 | } while (pos != kn); | |
fd7b9f7b TH |
1054 | } |
1055 | ||
1056 | /** | |
324a56e1 TH |
1057 | * kernfs_remove - remove a kernfs_node recursively |
1058 | * @kn: the kernfs_node to remove | |
fd7b9f7b | 1059 | * |
324a56e1 | 1060 | * Remove @kn along with all its subdirectories and files. |
fd7b9f7b | 1061 | */ |
324a56e1 | 1062 | void kernfs_remove(struct kernfs_node *kn) |
fd7b9f7b | 1063 | { |
988cd7af TH |
1064 | mutex_lock(&kernfs_mutex); |
1065 | __kernfs_remove(kn); | |
1066 | mutex_unlock(&kernfs_mutex); | |
fd7b9f7b TH |
1067 | } |
1068 | ||
6b0afc2a TH |
1069 | /** |
1070 | * kernfs_break_active_protection - break out of active protection | |
1071 | * @kn: the self kernfs_node | |
1072 | * | |
1073 | * The caller must be running off of a kernfs operation which is invoked | |
1074 | * with an active reference - e.g. one of kernfs_ops. Each invocation of | |
1075 | * this function must also be matched with an invocation of | |
1076 | * kernfs_unbreak_active_protection(). | |
1077 | * | |
1078 | * This function releases the active reference of @kn the caller is | |
1079 | * holding. Once this function is called, @kn may be removed at any point | |
1080 | * and the caller is solely responsible for ensuring that the objects it | |
1081 | * dereferences are accessible. | |
1082 | */ | |
1083 | void kernfs_break_active_protection(struct kernfs_node *kn) | |
1084 | { | |
1085 | /* | |
1086 | * Take out ourself out of the active ref dependency chain. If | |
1087 | * we're called without an active ref, lockdep will complain. | |
1088 | */ | |
1089 | kernfs_put_active(kn); | |
1090 | } | |
1091 | ||
1092 | /** | |
1093 | * kernfs_unbreak_active_protection - undo kernfs_break_active_protection() | |
1094 | * @kn: the self kernfs_node | |
1095 | * | |
1096 | * If kernfs_break_active_protection() was called, this function must be | |
1097 | * invoked before finishing the kernfs operation. Note that while this | |
1098 | * function restores the active reference, it doesn't and can't actually | |
1099 | * restore the active protection - @kn may already or be in the process of | |
1100 | * being removed. Once kernfs_break_active_protection() is invoked, that | |
1101 | * protection is irreversibly gone for the kernfs operation instance. | |
1102 | * | |
1103 | * While this function may be called at any point after | |
1104 | * kernfs_break_active_protection() is invoked, its most useful location | |
1105 | * would be right before the enclosing kernfs operation returns. | |
1106 | */ | |
1107 | void kernfs_unbreak_active_protection(struct kernfs_node *kn) | |
1108 | { | |
1109 | /* | |
1110 | * @kn->active could be in any state; however, the increment we do | |
1111 | * here will be undone as soon as the enclosing kernfs operation | |
1112 | * finishes and this temporary bump can't break anything. If @kn | |
1113 | * is alive, nothing changes. If @kn is being deactivated, the | |
1114 | * soon-to-follow put will either finish deactivation or restore | |
1115 | * deactivated state. If @kn is already removed, the temporary | |
1116 | * bump is guaranteed to be gone before @kn is released. | |
1117 | */ | |
1118 | atomic_inc(&kn->active); | |
1119 | if (kernfs_lockdep(kn)) | |
1120 | rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_); | |
1121 | } | |
1122 | ||
1123 | /** | |
1124 | * kernfs_remove_self - remove a kernfs_node from its own method | |
1125 | * @kn: the self kernfs_node to remove | |
1126 | * | |
1127 | * The caller must be running off of a kernfs operation which is invoked | |
1128 | * with an active reference - e.g. one of kernfs_ops. This can be used to | |
1129 | * implement a file operation which deletes itself. | |
1130 | * | |
1131 | * For example, the "delete" file for a sysfs device directory can be | |
1132 | * implemented by invoking kernfs_remove_self() on the "delete" file | |
1133 | * itself. This function breaks the circular dependency of trying to | |
1134 | * deactivate self while holding an active ref itself. It isn't necessary | |
1135 | * to modify the usual removal path to use kernfs_remove_self(). The | |
1136 | * "delete" implementation can simply invoke kernfs_remove_self() on self | |
1137 | * before proceeding with the usual removal path. kernfs will ignore later | |
1138 | * kernfs_remove() on self. | |
1139 | * | |
1140 | * kernfs_remove_self() can be called multiple times concurrently on the | |
1141 | * same kernfs_node. Only the first one actually performs removal and | |
1142 | * returns %true. All others will wait until the kernfs operation which | |
1143 | * won self-removal finishes and return %false. Note that the losers wait | |
1144 | * for the completion of not only the winning kernfs_remove_self() but also | |
1145 | * the whole kernfs_ops which won the arbitration. This can be used to | |
1146 | * guarantee, for example, all concurrent writes to a "delete" file to | |
1147 | * finish only after the whole operation is complete. | |
1148 | */ | |
1149 | bool kernfs_remove_self(struct kernfs_node *kn) | |
1150 | { | |
1151 | bool ret; | |
1152 | ||
1153 | mutex_lock(&kernfs_mutex); | |
1154 | kernfs_break_active_protection(kn); | |
1155 | ||
1156 | /* | |
1157 | * SUICIDAL is used to arbitrate among competing invocations. Only | |
1158 | * the first one will actually perform removal. When the removal | |
1159 | * is complete, SUICIDED is set and the active ref is restored | |
1160 | * while holding kernfs_mutex. The ones which lost arbitration | |
1161 | * waits for SUICDED && drained which can happen only after the | |
1162 | * enclosing kernfs operation which executed the winning instance | |
1163 | * of kernfs_remove_self() finished. | |
1164 | */ | |
1165 | if (!(kn->flags & KERNFS_SUICIDAL)) { | |
1166 | kn->flags |= KERNFS_SUICIDAL; | |
1167 | __kernfs_remove(kn); | |
1168 | kn->flags |= KERNFS_SUICIDED; | |
1169 | ret = true; | |
1170 | } else { | |
1171 | wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq; | |
1172 | DEFINE_WAIT(wait); | |
1173 | ||
1174 | while (true) { | |
1175 | prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE); | |
1176 | ||
1177 | if ((kn->flags & KERNFS_SUICIDED) && | |
1178 | atomic_read(&kn->active) == KN_DEACTIVATED_BIAS) | |
1179 | break; | |
1180 | ||
1181 | mutex_unlock(&kernfs_mutex); | |
1182 | schedule(); | |
1183 | mutex_lock(&kernfs_mutex); | |
1184 | } | |
1185 | finish_wait(waitq, &wait); | |
1186 | WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb)); | |
1187 | ret = false; | |
1188 | } | |
1189 | ||
1190 | /* | |
1191 | * This must be done while holding kernfs_mutex; otherwise, waiting | |
1192 | * for SUICIDED && deactivated could finish prematurely. | |
1193 | */ | |
1194 | kernfs_unbreak_active_protection(kn); | |
1195 | ||
1196 | mutex_unlock(&kernfs_mutex); | |
1197 | return ret; | |
1198 | } | |
1199 | ||
fd7b9f7b | 1200 | /** |
324a56e1 TH |
1201 | * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it |
1202 | * @parent: parent of the target | |
1203 | * @name: name of the kernfs_node to remove | |
1204 | * @ns: namespace tag of the kernfs_node to remove | |
fd7b9f7b | 1205 | * |
324a56e1 TH |
1206 | * Look for the kernfs_node with @name and @ns under @parent and remove it. |
1207 | * Returns 0 on success, -ENOENT if such entry doesn't exist. | |
fd7b9f7b | 1208 | */ |
324a56e1 | 1209 | int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name, |
fd7b9f7b TH |
1210 | const void *ns) |
1211 | { | |
324a56e1 | 1212 | struct kernfs_node *kn; |
fd7b9f7b | 1213 | |
324a56e1 | 1214 | if (!parent) { |
c637b8ac | 1215 | WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n", |
fd7b9f7b TH |
1216 | name); |
1217 | return -ENOENT; | |
1218 | } | |
1219 | ||
988cd7af | 1220 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1221 | |
324a56e1 TH |
1222 | kn = kernfs_find_ns(parent, name, ns); |
1223 | if (kn) | |
988cd7af | 1224 | __kernfs_remove(kn); |
fd7b9f7b | 1225 | |
988cd7af | 1226 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b | 1227 | |
324a56e1 | 1228 | if (kn) |
fd7b9f7b TH |
1229 | return 0; |
1230 | else | |
1231 | return -ENOENT; | |
1232 | } | |
1233 | ||
1234 | /** | |
1235 | * kernfs_rename_ns - move and rename a kernfs_node | |
324a56e1 | 1236 | * @kn: target node |
fd7b9f7b TH |
1237 | * @new_parent: new parent to put @sd under |
1238 | * @new_name: new name | |
1239 | * @new_ns: new namespace tag | |
1240 | */ | |
324a56e1 | 1241 | int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent, |
fd7b9f7b TH |
1242 | const char *new_name, const void *new_ns) |
1243 | { | |
3eef34ad TH |
1244 | struct kernfs_node *old_parent; |
1245 | const char *old_name = NULL; | |
fd7b9f7b TH |
1246 | int error; |
1247 | ||
3eef34ad TH |
1248 | /* can't move or rename root */ |
1249 | if (!kn->parent) | |
1250 | return -EINVAL; | |
1251 | ||
798c75a0 GKH |
1252 | mutex_lock(&kernfs_mutex); |
1253 | ||
d0ae3d43 | 1254 | error = -ENOENT; |
81c173cb | 1255 | if (!kernfs_active(kn) || !kernfs_active(new_parent)) |
d0ae3d43 TH |
1256 | goto out; |
1257 | ||
fd7b9f7b | 1258 | error = 0; |
adc5e8b5 TH |
1259 | if ((kn->parent == new_parent) && (kn->ns == new_ns) && |
1260 | (strcmp(kn->name, new_name) == 0)) | |
798c75a0 | 1261 | goto out; /* nothing to rename */ |
fd7b9f7b TH |
1262 | |
1263 | error = -EEXIST; | |
1264 | if (kernfs_find_ns(new_parent, new_name, new_ns)) | |
798c75a0 | 1265 | goto out; |
fd7b9f7b | 1266 | |
324a56e1 | 1267 | /* rename kernfs_node */ |
adc5e8b5 | 1268 | if (strcmp(kn->name, new_name) != 0) { |
fd7b9f7b TH |
1269 | error = -ENOMEM; |
1270 | new_name = kstrdup(new_name, GFP_KERNEL); | |
1271 | if (!new_name) | |
798c75a0 | 1272 | goto out; |
3eef34ad TH |
1273 | } else { |
1274 | new_name = NULL; | |
fd7b9f7b TH |
1275 | } |
1276 | ||
1277 | /* | |
1278 | * Move to the appropriate place in the appropriate directories rbtree. | |
1279 | */ | |
c637b8ac | 1280 | kernfs_unlink_sibling(kn); |
fd7b9f7b | 1281 | kernfs_get(new_parent); |
3eef34ad TH |
1282 | |
1283 | /* rename_lock protects ->parent and ->name accessors */ | |
1284 | spin_lock_irq(&kernfs_rename_lock); | |
1285 | ||
1286 | old_parent = kn->parent; | |
adc5e8b5 | 1287 | kn->parent = new_parent; |
3eef34ad TH |
1288 | |
1289 | kn->ns = new_ns; | |
1290 | if (new_name) { | |
1291 | if (!(kn->flags & KERNFS_STATIC_NAME)) | |
1292 | old_name = kn->name; | |
1293 | kn->flags &= ~KERNFS_STATIC_NAME; | |
1294 | kn->name = new_name; | |
1295 | } | |
1296 | ||
1297 | spin_unlock_irq(&kernfs_rename_lock); | |
1298 | ||
9561a896 | 1299 | kn->hash = kernfs_name_hash(kn->name, kn->ns); |
c637b8ac | 1300 | kernfs_link_sibling(kn); |
fd7b9f7b | 1301 | |
3eef34ad TH |
1302 | kernfs_put(old_parent); |
1303 | kfree(old_name); | |
1304 | ||
fd7b9f7b | 1305 | error = 0; |
798c75a0 | 1306 | out: |
a797bfc3 | 1307 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1308 | return error; |
1309 | } | |
1310 | ||
fd7b9f7b | 1311 | /* Relationship between s_mode and the DT_xxx types */ |
324a56e1 | 1312 | static inline unsigned char dt_type(struct kernfs_node *kn) |
fd7b9f7b | 1313 | { |
adc5e8b5 | 1314 | return (kn->mode >> 12) & 15; |
fd7b9f7b TH |
1315 | } |
1316 | ||
c637b8ac | 1317 | static int kernfs_dir_fop_release(struct inode *inode, struct file *filp) |
fd7b9f7b TH |
1318 | { |
1319 | kernfs_put(filp->private_data); | |
1320 | return 0; | |
1321 | } | |
1322 | ||
c637b8ac | 1323 | static struct kernfs_node *kernfs_dir_pos(const void *ns, |
324a56e1 | 1324 | struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos) |
fd7b9f7b TH |
1325 | { |
1326 | if (pos) { | |
81c173cb | 1327 | int valid = kernfs_active(pos) && |
798c75a0 | 1328 | pos->parent == parent && hash == pos->hash; |
fd7b9f7b TH |
1329 | kernfs_put(pos); |
1330 | if (!valid) | |
1331 | pos = NULL; | |
1332 | } | |
1333 | if (!pos && (hash > 1) && (hash < INT_MAX)) { | |
adc5e8b5 | 1334 | struct rb_node *node = parent->dir.children.rb_node; |
fd7b9f7b | 1335 | while (node) { |
324a56e1 | 1336 | pos = rb_to_kn(node); |
fd7b9f7b | 1337 | |
adc5e8b5 | 1338 | if (hash < pos->hash) |
fd7b9f7b | 1339 | node = node->rb_left; |
adc5e8b5 | 1340 | else if (hash > pos->hash) |
fd7b9f7b TH |
1341 | node = node->rb_right; |
1342 | else | |
1343 | break; | |
1344 | } | |
1345 | } | |
b9c9dad0 TH |
1346 | /* Skip over entries which are dying/dead or in the wrong namespace */ |
1347 | while (pos && (!kernfs_active(pos) || pos->ns != ns)) { | |
adc5e8b5 | 1348 | struct rb_node *node = rb_next(&pos->rb); |
fd7b9f7b TH |
1349 | if (!node) |
1350 | pos = NULL; | |
1351 | else | |
324a56e1 | 1352 | pos = rb_to_kn(node); |
fd7b9f7b TH |
1353 | } |
1354 | return pos; | |
1355 | } | |
1356 | ||
c637b8ac | 1357 | static struct kernfs_node *kernfs_dir_next_pos(const void *ns, |
324a56e1 | 1358 | struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos) |
fd7b9f7b | 1359 | { |
c637b8ac | 1360 | pos = kernfs_dir_pos(ns, parent, ino, pos); |
b9c9dad0 | 1361 | if (pos) { |
fd7b9f7b | 1362 | do { |
adc5e8b5 | 1363 | struct rb_node *node = rb_next(&pos->rb); |
fd7b9f7b TH |
1364 | if (!node) |
1365 | pos = NULL; | |
1366 | else | |
324a56e1 | 1367 | pos = rb_to_kn(node); |
b9c9dad0 TH |
1368 | } while (pos && (!kernfs_active(pos) || pos->ns != ns)); |
1369 | } | |
fd7b9f7b TH |
1370 | return pos; |
1371 | } | |
1372 | ||
c637b8ac | 1373 | static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx) |
fd7b9f7b TH |
1374 | { |
1375 | struct dentry *dentry = file->f_path.dentry; | |
324a56e1 TH |
1376 | struct kernfs_node *parent = dentry->d_fsdata; |
1377 | struct kernfs_node *pos = file->private_data; | |
fd7b9f7b TH |
1378 | const void *ns = NULL; |
1379 | ||
1380 | if (!dir_emit_dots(file, ctx)) | |
1381 | return 0; | |
a797bfc3 | 1382 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1383 | |
324a56e1 | 1384 | if (kernfs_ns_enabled(parent)) |
c525aadd | 1385 | ns = kernfs_info(dentry->d_sb)->ns; |
fd7b9f7b | 1386 | |
c637b8ac | 1387 | for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos); |
fd7b9f7b | 1388 | pos; |
c637b8ac | 1389 | pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) { |
adc5e8b5 | 1390 | const char *name = pos->name; |
fd7b9f7b TH |
1391 | unsigned int type = dt_type(pos); |
1392 | int len = strlen(name); | |
adc5e8b5 | 1393 | ino_t ino = pos->ino; |
fd7b9f7b | 1394 | |
adc5e8b5 | 1395 | ctx->pos = pos->hash; |
fd7b9f7b TH |
1396 | file->private_data = pos; |
1397 | kernfs_get(pos); | |
1398 | ||
a797bfc3 | 1399 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1400 | if (!dir_emit(ctx, name, len, ino, type)) |
1401 | return 0; | |
a797bfc3 | 1402 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1403 | } |
a797bfc3 | 1404 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1405 | file->private_data = NULL; |
1406 | ctx->pos = INT_MAX; | |
1407 | return 0; | |
1408 | } | |
1409 | ||
c637b8ac TH |
1410 | static loff_t kernfs_dir_fop_llseek(struct file *file, loff_t offset, |
1411 | int whence) | |
fd7b9f7b TH |
1412 | { |
1413 | struct inode *inode = file_inode(file); | |
1414 | loff_t ret; | |
1415 | ||
1416 | mutex_lock(&inode->i_mutex); | |
1417 | ret = generic_file_llseek(file, offset, whence); | |
1418 | mutex_unlock(&inode->i_mutex); | |
1419 | ||
1420 | return ret; | |
1421 | } | |
1422 | ||
a797bfc3 | 1423 | const struct file_operations kernfs_dir_fops = { |
fd7b9f7b | 1424 | .read = generic_read_dir, |
c637b8ac TH |
1425 | .iterate = kernfs_fop_readdir, |
1426 | .release = kernfs_dir_fop_release, | |
1427 | .llseek = kernfs_dir_fop_llseek, | |
fd7b9f7b | 1428 | }; |